EAGER: Towards A Lightweight and Personalized Implicit Authentication System with Adaptive Sensing

点击翻译按钮获取中文摘要

Hard biometrics-based authentication methods such as face recognition, touch ID and iris scan require explicit action from users and expensive hardware. In contrast, behaviors are soft biometrics suitable for implicit and cost-efficient user authentication. This authentication method can lead to improved security by reducing the risk of users bypassing ill-designed authentication. It also relieves users from the burden of memorizing passwords, revolutionarily enhancing user experience. However, the fact that behaviors can change with many factors such as age, mood, and environment renders it more challenging to develop systems that depend on behaviors. This research project focuses on designing the first implicit authentication system that automatically selects dynamic sets of activities for user behavior extraction, and building a prototype authentication system. To ensure practical deployment, lightweight computation and adaptive sensing are integrated into the system, making it energy-efficient for popular energy-constraint devices.The proposed research project is an exploratory project in its early stages. One of the key contributions and unique aspects of the proposed research is that it sets the foundation for a cutting-edge engineering system that is still in its infancy and whose complexity extends far beyond a few novel algorithms. The most hindering factor of behavior-based implicit authentication systems is the difficulties in handling dynamic behavior changes that make the system unreliable. This project takes a drastically different approach by selecting the most suitable set of behaviors over time from available behaviors, and is thus a general framework that is device-independent. Three research tasks will be carried out: Modeling and selecting the dynamically changing behaviors unique for each person based on probabilistic models; Developing an energy-efficient and practical client side by solving challenging issues such as differentiating legitimate users behavior deviation from illegitimate users; Building a prototype authentication system to handle authentication failures and enhance user experience. The proposed research is potentially transformative. Exploring behaviors, activity patterns and habits serves as the basis for dynamic, intelligent, versatile, seamless and user-friendly designs, transforming modern computer systems and applications into our smarter assistants, educators, health watchers, security vaults, entertainers, play mates and many more. The proposed project has developed viable methods to capture users evolving behaviors, dynamically adjust the models, and abstract system components so that the generic system can be easily reusable and extensible. The research findings will contribute to advancing the state-of-the-art research, and stimulating the wide adoption of implicit authentication systems in various domains including cybersecurity, healthcare and fitness, mobile social networks, and e-commerce.